Over the past couple of decades we’ve witnessed a whirlwind of cultural changes in the music industry, but also major changes in terms of how we find and listen to music. And there’s arguably one entity that has contributed to these shifts more than any other: Spotify — which was founded 20 years ago today (April 23). Feel old yet? I sure do.

For many music lovers out there, myself included, Spotify was their introduction to music streaming, and over the last 20 years it’s climbed to the top of the ladder, amassing over 750 million users and cementing its position as one of the best music streaming services — and in the eyes of many, the daddy of them all.

NASA announced that it will launch the Nancy Grace Roman space telescope into orbit in September 2026, eight months ahead of schedule. The new space telescope is expected to deliver 20,000 terabytes of data to astronomers over the course of its life.

That will add to 57 gigabytes of breath-taking imagery downlinked daily from the James Webb Space Telescope, which began its work in 2021, and the start of a survey later this year by the Vera C. Rubin Observatory in the mountains of Chile, which is expected to gather 20 terabytes of data each night.

For comparison, the Hubble Space telescope, once the gold standard, delivers just 1 to 2 gigabytes of sensor readings each day. It’s been a while since all those readings were pored over by hand, but like everyone else with a pile of data, astronomers are now turning to GPUs to solve their problems.

Brant Robertson, a UC Santa Cruz astrophysicist, has had a front-row seat to this step change in science while supporting or using data from these missions. Robertson has spent the past 15 years working with Nvidia to apply GPUs to the problems of understanding space, first through advanced simulations testing theories about supernnova explosions, and now developing the tools to analyze a torrent of data from the newest observatories.

“There’s been this evolution [from] looking at a few objects, to doing CPU-based analyses on large scales of the data set, to then doing GPU-accelerated versions of those same analyses,” he told TechCrunch.

Robertson and then-graduate student Ryan Hausen developed a deep learning model called Morpheus that can pore over large data sets and identify galaxies. Their early AI analysis of Webb data identified a surprising number of a specific type of disc galaxies and added a new wrinkle to theories about the development of our universe.

Now Morpheus is changing with the times: Robertson is switching its architecture from convolutional neural networks to the transformers behind the rise of large language models. That will result in the model being able to analyze several times the area than it can currently, speeding up its work.

Robertson is also working on generative AI models trained on space telescope data to improve the quality of observations collected by ground telescopes, which are distorted by Earth’s atmosphere. Despite advances in rocketry, it’s still hard to get an 8 meter mirror into orbit, so using software to improve Rubin’s observations is the next best thing.

But he’s still feeling the pressure of global demand for GPU access. Robertson has used the National Science Foundation to build a GPU cluster at UC Santa Cruz, but it is becoming outdated even as more researchers want to apply compute-intensive techniques to their work. The Trump administration proposed cutting the NSF’s budget by 50% in its current budget request.

“People want to do these AI, ML analyses, and GPUs are really the way to do that,” Robertson said. “You have to be entrepreneurial…especially when you’re working kind of at the edge of where the technology is. Universities are very risk averse because they just have constrained resources, so you have to go out and show them that, ‘look, this is where we’re going as a field.’”

There’s little point in setting up your own shed-based clean room for semiconductor purposes if you don’t try to do something practical with it. Something like responding to the RAMpocalypse by trying to make your own RAM, for example.

After all, what could be so hard about etching the same repeating structures over and over? In a recent video, [Dr. Semiconductor]’s experience doing exactly this are detailed, with actual DRAM resulting at the end.

We covered the construction of the clean room shed previously, which should provide at least the basic conditions to produce semiconductors without worrying about contaminating dies. From here the process is reminiscent of etching PCBs, with a prepared surface coated with photoresist. Using UV exposure through a mask, the pattern is etched into the photoresist and from there the pattern is subsequently etched into the wafer’s surface.

With the patterns formed, the next step is doping of the silicon in order to create the active structures, i.e. the transistors and capacitors. Doping can be done in a variety of ways, with ion implantation being the industry standard method, but a bit too expensive and bulky for a shed fab. Instead a spin-on-glass method was used. After this the remaining functional structures can be built up.

If anyone was expecting to see a DDR5 DRAM die pop out at the end, they’re bound to be disappointed. The target here was to create a 5×4 array of DRAM cells, for a dizzying 20 bits. Still, the fact that it’s possible to DIY DRAM like this at home is already pretty awesome, with clearly plenty of room to push it towards and past fabrication nodes of the 1990s and beyond.

Although the produced DRAM cells have fairly leaky capacitors, they’re good enough for their purpose, and the plan is to scale up to a large DRAM array from here. Whether the DRAM control logic will also be implemented in hardware like this remains to be seen, but the video’s ending makes it clear that the goal is to attach it to a PC somehow.

I bought a Ring Doorbell a while back and it’s been incredibly useful. When I’m working, I can see if I should answer the door or not. I have added peace of mind at night, and I can monitor the outside of my home any time I’m on vacation.

Now, the Ring Battery Doorbell has dropped to $59.99 (was $99.99) at Amazon.

Previously, the Ring Battery Doorbell has been a little cheaper at $49.99, but 40% off is still a great deal. That lower price was back during Black Friday, so don’t count on it being cheaper than this any time soon.

The Ring Battery Doorbell lets you easily see who’s calling at your home, no matter where you are, and you can talk to them, too. It’s a good security measure and a great way to avoid missing parcels. That’s a lot of useful features for such a low price.

Today’s best video doorbell deal

Ring is a brand that often features among the best video doorbells. While the Ring Battery Doorbell isn’t the most feature-packed option, it’s very good value for money.

It’s the one I use for my home. It means I get real-time alerts to my phone any time motion is detected, such as if a delivery person is calling or a friend has knocked on the door. Its bell chime reverberates through the house thanks to my many Alexa speakers, which easily pair with it.

I subscribe to Ring’s premium service so I can scroll through the history to see what’s happened. It isn’t entirely essential to subscribe if you just want a regular but smarter doorbell, but I’d recommend it. It’s very useful. It also works with more advanced models like the Ring Video Doorbell Plus. By committing, it can even replicate the best home security cameras pretty well.

Battery life is also pretty good. I find I only have to recharge it every few weeks, and that’s mostly because it detects a lot of motion due to me living on a busy street and leaving the motion zones quite wide. For many people, it’ll last a lot longer. It’s a nice hands-off solution.

There are other Ring doorbell deals around if you’re considering a pricier model or a wired solution. While we’re talking about things that have improved my life massively, there are lots of air fryer deals around too.

Once upon a time in Europe, television remote controls had a magic teletext button. Years before the internet stole into homes, pressing that button brought up teletext digital information services with hundreds of constantly updated pages. Living in Ireland in the 1980s and ’90s, my family accessed the national teletext service—Aertel—multiple times a day for weather and news bulletins, as well as things like TV program guides and updates on airport flight arrivals.

It was an elegant system: fast, low bandwidth, unaffected by user load, and delivering readable text even on analog television screens. So when I recently saw it was the 40th anniversary of Aertel’s test transmissions, it reactivated a thought that had been rolling around in my head for years. Could I make a ham-radio version of teletext?

What is Teletext?

First developed in the United Kingdom and rolled out to the public by the BBC under the name Ceefax, teletext exploited a quirk of analog television signals. These signals transmitted video frames as lines of luminosity and color, plus some additional blank lines that weren’t displayed. Teletext piggybacked a digital signal onto these spares, transmitting a carousel of pages over time. Using their remotes, viewers typed in the three-digit code of the page they wanted. Generally within a few seconds, the carousel would cycle around and display the desired page.

Teletext created unusually legible text in the 8-bit era by enlarging alphanumeric characters and interpolating new pixels by looking for existing pixels touching diagonally, and adding whitespace between characters. Graphic characters were not interpolated, and featured blocky chunks known as sixels for their 2-by-3 arrangement. My modern recreation uses the open-source font Bedstead, which replicates the look of teletext, including the graphics characters. James Provost

Teletext is composed of characters that can be one of eight colors. Control codes in the character stream select colors and can also produce effects like flashing text and double-height characters. The text’s legibility was better than most computers could manage at the time, thanks to the SAA5050 character-generator chip at the heart of teletext. Although characters are internally stored on this chip in 6-by-10-pixel cells—fewer pixels than the typical 8-by-8-pixel cell used in 1980s home computers—the SAA5050 interpolates additional pixels for alphanumeric characters on the fly, making the effective resolution 10 by 18 pixels. The trade-off is very low-resolution graphics, comprising characters that use a 2-by-3 set of blocky pixels.

Teletext screens use a 40-by-24-character grid. This means that a kilobyte of memory can store a full page of multicolor text, half the memory required for a similar amount of text on, for example, the Commodore 64. The BBC Microcomputer took advantage of this by putting an SAA5050 on its motherboard, which could be accessed in one of the computer’s graphics modes. Despite the crude graphics, some educational games used this mode, most notably Granny’s Garden, which filled the same cultural niche among British schoolchildren that The Oregon Trail did for their U.S. counterparts.

By the 2010s, most teletext services had ceased broadcasting. But teletext is still remembered fondly by many, and enthusiasts are keeping it alive, recovering and archiving old content, running internet-based services with current newsfeeds, and developing systems that make it possible to create and display teletext with modern TVs.

Putting Teletext Back on the Air

I wanted to do something a little different. Inspired by how the BBC Micro co-opted teletext for its own purposes, I thought it might make a great radio protocol. In particular I thought it could be a digital counterpart to slow-scan television (SSTV).

SSTV is an analog method of transmitting pictures, typically including banners with ham-radio call signs and other messages. SSTV is fun, but, true to its name, it’s slow—the most popular protocols take a little under 2 minutes to send an image—and it can be tricky to get a complete picture with legible text. For that reason, SSTV images are often broadcast multiple times.

Teletext is still remembered fondly by many.

I decided to send the teletext using the AX.25 protocol, which encodes ones and zeros as audible tones. For VHF and UHF transmissions at a rate of 1,200 baud, it would take 11 seconds to send one teletext screen. Over HF bands, AX.25 data is normally sent at 300 baud, which would result in a still-acceptable 44 seconds per screen. When a teletext page is sent repeatedly, any missed or corrupted rows are filled in with new ones. So in a little over 2 minutes, I could send a screen three times over HF, and the receiver would automatically combine the data. I also wanted to build the system in Python for portability, with an editor for creating pages, an AX.25 encoder and decoder, and a monitor for displaying received images.

The reason why I hadn’t done this before was because it requires digesting the details of the AX.25 standard and teletext’s official spec, and then translating them into a suite of software, which I never seemed to have the time to do. So I tried an experiment within an experiment, and turned to vibe coding.

Despite the popularity of vibe coding with developers, I have reservations. Even if concerns about AI slop, the environment, and memory hoarding were not on the table, I would still worry about the reliance on centralized systems that vibe coding brings. The whole point of a DIY project is to, well, do it yourself. A DIY project lets you craft things for your own purposes, not just operate within someone else’s profit margins and policies.

Still, criticizing a technology from afar isn’t ideal, so I directed Anthropic’s Claude toward the AX.25 and teletext specs and told it what I wanted. After about 250,000 to 300,000 tokens and several nights of back and forth about bugs and features, I had the complete system running without writing a single line of code. Being honest with myself, I doubt this system—which I’m calling Spectel—would ever have come about without vibe coding.

But I didn’t learn anything new about how teletext works, and only a little bit more about AX.25. Updates are contingent on my paying Anthropic’s fees. So I remain deeply ambivalent about vibe coding. And one final test remains in any case: trying Spectel out on HF bands. Of course, that means I’ll need willing partners out in the ether. So if you’re a ham who’d like to help out, let me know in the comments below!